The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Simulators >> AMS Simulators >> accuracy in transient simulations
https://designers-guide.org/forum/YaBB.pl?num=1129214370

Message started by vivkr on Oct 13^th, 2005, 7:39am

Title: accuracy in transient simulations
Post by vivkr on Oct 13^th, 2005, 7:39am

I am trying to simulate a very highly accurate circuit, and am not very sure if the errors I see are due to the circuit or the simulator. My circuit is a switched-capacitor filter.

I tightened all the accuracy parameters considerably (reltol, abstol) and set relref to "pointlocal". I am also using gear2 integration method to prevent trapezoidal ringing.

However, I see the following dialog

"Convergence difficulties resulted in error requirements being unsatisfied".

I don't know how much simulator error I have now, and hence it is very difficult to get a reliable estimate for the error in the charge-transfer process.

I looked up the guidelines in the "Designer's Guide to Spice & Spectre" and try to use as many of those as I can to achieve better convergence. I cannot use "gmin" as I have a charge-transfer going on, but I use "cmin".

Could someone suggest some tips for improving convergence and accuracy in Spectre transient analysis?

Many thanks and Best Regards
Vivek

Title: Re: accuracy in transient simulations
Post by Geoffrey_Coram on Nov 14^th, 2005, 8:32am

Have you perhaps over-tightened your tolerances? It's possible that the values you set aren't possible to achieve given the finite numerical precision of the computer. If you have 1A out of one device and into another, you shouldn't expect to get better than 1e-15A for KCL at the node; if you set reltol and abstol to 1e-15, you might never get convergence.

Title: Re: accuracy in transient simulations
Post by sheldon on Nov 18^th, 2005, 9:31pm

Vivkr,

Is the circuit entirely transistor level or is it mixed-blocks of
transistor and behavior models? Usually this warning is telling
you that you have a badly behaved behavioral model. For
example, an output from a conditional without a transition
filter. I believe that Geoffrey is correct your simulator tolerances
maybe too tight and it would be it would be better to look at the
device and behavioral models for potential issues.

Best Regards,

Sheldon

Title: Re: accuracy in transient simulations
Post by Ken Kundert on Nov 18^th, 2005, 10:20pm

You have set the tolerances too tight. You should start by setting them all back to their default values, and then follow the guidelines given in Section 4.4.4 (page 224) of "The Designer's Guide to SPICE and Spectre".

-Ken